July 10, 2025
About the Company
Nanya is a small Taiwanese DRAM manufacturer. They produce DRAM on trailing-edge process nodes (licensed from Micron) but are developing their own more advanced, technology. Nanya has about a 1.6% share of the DRAM market, a distant fourth, behind Samsung, Hynix, and Micron. The top three companies control a combined 96.4% of the DRAM market.
I don’t own any shares of Nanya and would not buy any, because they lack the scale and the technology to be competitive in DRAM. I keep an eye on their quarterly releases as a data point on the DRAM market.
Market Commentary and Q2 2025 Performance
Ten quarters have passed since Nanya stopped including the time trend of their gross margin with their earnings. While gross margin deteriorated further (it fell 560 bps, to negative 20.6%) revenue surged by a staggering 46.4%. The company doesn’t release full financial statements with their results, but they must have drawn down inventory to ship this number of bits. The company has not shipped this volume of bits in at least three years. Even with that deluge of supply, the markets Nanya serves had the capacity to absorb the bits without collapsing prices. Nanya’s ASPs only declined by 5%. This follows a 1% drop the quarter prior and an 11% sequential fall two quarters ago. Two quarters ago, I thought it was over for Nanya. I thought the DRAM market had slid back into oversupply, a situation caused by the entrance of CXMT into the low end of the DRAM supply pool. But it appears all the bits being soaked up by HBM are finally having an effect on the non-AI DRAM market. DDR4 spot prices are surging and Nanya’s customers are clamoring for supply. Still, the company’s operating margin only improved by 110 bps and stands at negative 42.8%. That is not sustainable. It leads to a net margin of negative 39.0%. This is 1200 bps worse than last quarter, but the reason is a non-operating expense. While results aren’t as bad as 40 cents of every dollar of revenue being lost, they are still terrible. Nanya is treading water and still sinking.
Bit Shipments and Capital Expenditures
The company has spent NT$9.3B of capital expenditures in the first half of calendar 2025. That is out of a total of ~NT$19.6B approved by the Board back in February. If they execute on this spending budget, it will almost match the NT$20.7B the company spent in this category back in 2022. In that year, bit shipments were still down 20%. That was followed by a mid-single digit % decline in 2023 and no bit growth in 2024. Where did the CapEx go? It went to technology node transitions, R&D, and cleanroom space. They are spending too little money though, because bits still need to grow as tech nodes progress. Otherwise, the investment doesn’t pay off. Nanya is slowly eating itself, because it can’t afford to spend what it needs to spend to get to the scale required to succeed. Still, they are finally going to see some bit growth from the investments of the last three years. As of this announcement, the company is forecasting >40% bit growth in 2025 over 2024. Last quarter they said it would be up >30% this year. In the announcement before that, they said it would be up >20% this year. I think what happened is they got the capital expenditure increase approved in February and then have found better ways to spend that money, leading to increasing bit shipment commitments for 2025. To meet this forecast of >40% bit growth year-over-year, Nanya needs to ship more bits in each of the next two quarters than they did in Q2, by about 10%.
DRAM Market Outlook for Q3 of 2025
Demand for DRAM products continues to be driven by the data center segment. The market for non-AI products bottomed out sometime in the second calendar quarter of 2025. The company believes the second half of 2025 will see relative stability. DDR4 and LPDDR4 supplies have been reduced, leading to improving pricing for those products in the spot market.
- · Server Market: Demand from AI server construction continues to undergird the DRAM market.
- · Mobile Market: Hopefully the smartphone upgrade cycle will happen soon.
- · PC Market: Hopefully AI PCs become a thing and that causes people to buy more PCs. No sign of that yet.
- · Consumer Market: Use of DDR4 and LPDDR4 are declining in PCs and smartphones. Those architectures are being adopted by consumer applications now.
Analyst Call
Prepared Remarks
- · Pei-ing Lee, the CEO, started with some opening comments. He said the non-AI DRAM market had been in a cyclical downturn until May of this year. They are seeing recovery in June. Therefore, most of the current Q2 occurred when conditions had not yet improved.
- · In April and May, ASPs declined. They started to rise in June.
- · The transition to DDR5 has now led to a squeeze in DDR4 supply as bits are transitioned away from the older architecture.
Analyst Q&A
- · Management is “very confident” their gross margin will turn positive in the third quarter. That is really surprising given how far from zero their GM was in Q2.
- · Most of their 1-b node capacity is currently making both D/LPDDR4 and D/LPDDR5 products. By the end of the year, they expect more than 30% of their output to be on the 1-b node.
- · Because of longer contracts the company had already entered in to, the weak pricing in April and May had an outsized effect on their Q2 results. The improving market has seen customers sign more longer-term contracts with better pricing.
- · They don’t run any 1-a node now. Their capacity is all 20nm (~70%) and 1-b (~30%). In Q4 of this year, the two nodes, measured by bit output, will be about even.
- · The CEO said their 1-b node doesn’t have much of a cost advantage over their prior node. He believes this is true for most suppliers. This is false. The main reason for node transitions is cost reduction. I think they have some structural problem(s) that are preventing them from realizing those cost advantages, and he is justifying that to himself. He said the main advantage is being able to make higher density products. This was really surprising to hear, such a profound misunderstanding of the industry from a memory company leader.
- · In the second half of this year, DDR4 and LPDDR4 will still be more than 50% of their output. The company still makes a lot of DDR3 and low power. In the most recent quarter, those products were a larger fraction than DDR5 for them.
- · Most of their sales are for monthly and quarterly contracts. Their spot exposure is low.
- · In Q2, DDR5 revenue contribution was in the teens percent of total. Dr. Lee would not say what this fraction will be in the second half of 2025.
- · Total wafer capacity for the company will be similar in 2026 as it is in 2025.
Summary
I’ll close by commenting on Nanya’s situation as a company and then on what their results reveal about the markets they sell into. The company is subscale. They are investing in more advanced process nodes but can’t afford to do so at a scale that makes those transitions economical. The substantial investments they have made recently and are making this year are finally translating into more bit output. In each of the last three quarters they have raised their bit output forecast for calendar 2025. It is now up to a 40% sequential rise over 2024. While a small change in the scale of the overall DRAM market (Nanya’s market share is less than 3%,) it is in the right direction for oversupply. Nanya’s markets finally turned sharply upward in June. That is, DDR4 and older products slid into shortage. This tells us the effect of the three major DRAM players shifting capacity over to HBM and taking those bits away from non-AI memory has finally moved some DRAM markets into shortage. The big players have also been moving as fast as they can to DDR5 from DDR4. Those two actions were destined to move the DDR4 an LPDDR4 markets into undersupply eventually. That has happened as of the April to June period of this year.
-S. Hughes (short MU)